Hey PN, I'm the engineer at Basho working on Ring Resizing, primarily. I've left some comments inline.
On Wed, Sep 18, 2013 at 12:46 AM, <pseudo-n...@laposte.net> wrote: > > Dear all, > After a quick glimpse at Riak, I must say I like it: DHT with masterless > design and gossiping protocol, that is appealing. More than maintainability > or configurability, I am interested in scalability. I have looked at the > node adding process for fine scaling, but for scaling big (over the long > term) the necessity of resizing the ring (in order to double the number of > partitions) seems unavoidable to me. > This is certainly one of the use cases the feature is meant to address. However, we suggest that it is your "last resort" scaling feature -- if your cluster can still handle adding nodes that's a better option. But at some point, yes, doubling or quadrupling the ring size can be useful, especially if you started with a small cluster. Its, not useful, however, for e.g. doubling the number of partitions during peak hours for more concurrency and then shrinking back down again. > - is this feature officially released as part of 1.4? > The feature was released as experimental in Riak 1.4. The move out of experimental is slated for the next release. > - if yes, is it a stable feature, is it ready for production? if not, is > it still under ongoing development? > - if it was released as unstable, what is your experience with it? > For open source, using this feature w/ the 1.4 release is not suggested for production use. That should change with the next release. This is why there is no documentation. We are currently working with a few customers, who have clusters suited for resizing, at various levels of testing/production. It is currently under further development. Immediate work involves bug fixes and minor improvements. Longer term improvements will focus on reducing the overhead/cost of the operation. > - in your opinion, can the problem be simplified if we only consider > scaling out by doubling the ring size? > The problem is pretty much the same whether you are expanding by a factor of 2 or 4 (same for shrinking). The larger the growth/shrink factor the slower the operation will be. Also, independently of ring resizing, what do you consider the current > limits of ring_creation_size? I have seen mentioned that 2048 and higher > could importantly increase the volume of gossiping, what is your experience > with it? > I hope you will be kind enough to provide me with some enlightenments. > Kind regards. > PN > There we some changes recently to riak_core to improve its handling of large rings [1]. The benchmarks there tested up to 16384, however, i believe that was for illustration purposes. Most large clusters we see use 1024. The theoretical maximum most likely depends on many things. Cheers, Jordan > > > > Une messagerie gratuite, garantie à vie et des services en plus, ça vous > tente ? > Je crée ma boîte mail www.laposte.net > > _______________________________________________ > riak-users mailing list > riak-users@lists.basho.com > http://lists.basho.com/mailman/listinfo/riak-users_lists.basho.com >
_______________________________________________ riak-users mailing list riak-users@lists.basho.com http://lists.basho.com/mailman/listinfo/riak-users_lists.basho.com